Trump signs AI security executive order; 30-day voluntary prerelease model review

President Trump signed an executive order on June 2 titled 'Promoting Advanced Artificial Intelligence Innovation and Security' that directs federal agencies to establish a framework for voluntary pre-release government access to frontier AI models. Developers of advanced models would submit their systems to the government for up to 30 days of testing before releasing them more broadly to trusted partners. The order does not create a mandatory licensing or preclearance regime but emphasizes collaborative public-private partnership. The shift reflects administration recognition that frontier models with advanced cyber capabilities (like Anthropic's Mythos and OpenAI's GPT-5.5-Cyber) pose novel national security and critical-infrastructure risks.

The order creates an 'AI cybersecurity clearinghouse' overseen by Treasury, National Cyber Director, War, and Homeland Security, tasked with collecting and sharing vulnerability data and threat intelligence. Agencies have 30–60 days to prioritize cyber defense of Department of War, civilian federal, and critical-infrastructure systems by deploying AI-enabled defensive tools. The order also prioritizes criminal enforcement against AI-enabled cyberattacks and directs development of classified benchmarks to assess frontier models' cyber capabilities, with results informing future engagement with developers.

The order signals a notable evolution from the Trump administration's January 2025 'Removing Barriers' EO, which focused on deregulation. The change came after Anthropic's May 2026 expansion of Mythos (a model purpose-built to identify software vulnerabilities) from ~50 to ~200 organizations, plus lobbying from Musk, Sacks, and Zuckerberg against a stricter 90-day review timeline. The final 30-day window reflects compromise between national-security and innovation factions within the administration.

For architects: the order signals government interest in frontier AI cyber capabilities and implicit acknowledgment that model diffusion timelines now carry infrastructure risk. Although framed as voluntary, participation may become de facto mandatory due to national-security scrutiny and reputational cost. Frontier model developers should prepare for classified benchmarking and consider disclosure terms (what gets shared, with whom, under what retention/distribution rules). Monitor CISA guidance on critical-infrastructure AI tooling access and watch how 'frontier model' definitions emerge in forthcoming agency rulemaking. The framework sits somewhere between pre-release safety testing (Biden-era) and national-security vetting—neither light-touch nor heavyweight.